Face Generation¶

In this project, you'll use generative adversarial networks to generate new images of faces.

Get the Data¶

You'll be using two datasets in this project:

MNIST
CelebA

Since the celebA dataset is complex and you're doing GANs in a project for the first time, we want you to test your neural network on MNIST before CelebA. Running the GANs on MNIST will allow you to see how well your model trains sooner.

If you're using FloydHub, set data_dir to "/input" and use the FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

data_dir = './data'
!pip install matplotlib==2.0.2
# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'



"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)

Collecting matplotlib==2.0.2
  Using cached https://files.pythonhosted.org/packages/2c/69/362c0a5eca37225c7d451e6c57f07c4fee85a5b6cb3aedf16fe649f411ea/matplotlib-2.0.2-cp27-cp27mu-manylinux1_x86_64.whl
Requirement already satisfied: subprocess32 in /usr/local/lib/python2.7/dist-packages (from matplotlib==2.0.2)
Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=1.5.6 in /usr/local/lib/python2.7/dist-packages (from matplotlib==2.0.2)
Requirement already satisfied: pytz in /usr/local/lib/python2.7/dist-packages (from matplotlib==2.0.2)
Requirement already satisfied: six>=1.10 in /usr/local/lib/python2.7/dist-packages (from matplotlib==2.0.2)
Requirement already satisfied: functools32 in /usr/local/lib/python2.7/dist-packages (from matplotlib==2.0.2)
Requirement already satisfied: python-dateutil in /usr/local/lib/python2.7/dist-packages (from matplotlib==2.0.2)
Requirement already satisfied: cycler>=0.10 in /usr/local/lib/python2.7/dist-packages (from matplotlib==2.0.2)
Requirement already satisfied: numpy>=1.7.1 in /usr/local/lib/python2.7/dist-packages (from matplotlib==2.0.2)
Installing collected packages: matplotlib
  Found existing installation: matplotlib 2.1.0
    Uninstalling matplotlib-2.1.0:
Exception:
Traceback (most recent call last):
  File "/usr/local/lib/python2.7/dist-packages/pip/basecommand.py", line 215, in main
    status = self.run(options, args)
  File "/usr/local/lib/python2.7/dist-packages/pip/commands/install.py", line 342, in run
    prefix=options.prefix_path,
  File "/usr/local/lib/python2.7/dist-packages/pip/req/req_set.py", line 778, in install
    requirement.uninstall(auto_confirm=True)
  File "/usr/local/lib/python2.7/dist-packages/pip/req/req_install.py", line 754, in uninstall
    paths_to_remove.remove(auto_confirm)
  File "/usr/local/lib/python2.7/dist-packages/pip/req/req_uninstall.py", line 115, in remove
    renames(path, new_path)
  File "/usr/local/lib/python2.7/dist-packages/pip/utils/__init__.py", line 267, in renames
    shutil.move(old, new)
  File "/usr/lib/python2.7/shutil.py", line 303, in move
    os.unlink(src)
OSError: [Errno 13] Permission denied: '/usr/local/lib/python2.7/dist-packages/matplotlib-2.1.0-py2.7-nspkg.pth'
You are using pip version 9.0.1, however version 18.0 is available.
You should consider upgrading via the 'pip install --upgrade pip' command.
Found mnist Data
Found celeba Data

Explore the Data¶

MNIST¶

As you're aware, the MNIST dataset contains images of handwritten digits. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x7f312056a128>

CelebA¶

The CelebFaces Attributes Dataset (CelebA) dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x7f3120498978>

Preprocess the Data¶

Since the project's main focus is on building the GANs, we'll preprocess the data for you. The values of the MNIST and CelebA dataset will be in the range of -0.5 to 0.5 of 28x28 dimensional images. The CelebA images will be cropped to remove parts of the image that don't include a face, then resized down to 28x28.

The MNIST images are black and white images with a single color channel while the CelebA images have 3 color channels (RGB color channel).

Build the Neural Network¶

You'll build the components necessary to build a GANs by implementing the following functions below:

model_inputs
discriminator
generator
model_loss
model_opt
train

Check the Version of TensorFlow and Access to GPU¶

This will check to make sure you have the correct version of TensorFlow and access to a GPU

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.0.0
Default GPU Device: /gpu:0

Input¶

Implement the model_inputs function to create TF Placeholders for the Neural Network. It should create the following placeholders:

Real input images placeholder with rank 4 using image_width, image_height, and image_channels.
Z input placeholder with rank 2 using z_dim.
Learning rate placeholder with rank 0.

Return the placeholders in the following the tuple (tensor of real input images, tensor of z data)

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    # TODO: Implement Function
    inputs = tf.placeholder(tf.float32, [None, image_width, image_height, image_channels],name='inputs')
    z = tf.placeholder(tf.float32, [None,z_dim], name='z')
    learning_rate = tf.placeholder(tf.float32, None, name='learning_rate')
    
    return (inputs, z, learning_rate)


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

Tests Passed

Discriminator¶

Implement discriminator to create a discriminator neural network that discriminates on images. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "discriminator" to allow the variables to be reused. The function should return a tuple of (tensor output of the discriminator, tensor logits of the discriminator).

def discriminator(images, reuse=False,alpha=0.1):
    """
    Create the discriminator network
    :param images: Tensor of input image(s)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """
    
    with tf.variable_scope('discriminator',reuse=reuse):
        
        #input is (28,28,3) ((28,28,1) in Mnist)
        
        conv1 = tf.layers.conv2d(images,128,5,strides=2,padding='same')
        #conv1 = tf.layers.batch_normalization(conv1,training=True) *no normalization in first layer
        conv1 = tf.maximum(alpha * conv1, conv1)
        #shape is (14,14,128)
        
        conv2 = tf.layers.conv2d(conv1,256,5,strides=2,padding='same')
        conv2 = tf.layers.batch_normalization(conv2,training=True)
        conv2 = tf.layers.dropout(conv2,0.2)
        conv2 = tf.maximum(alpha * conv2, conv2)
        #shape is (7,7,256)
        
        conv3 = tf.layers.conv2d(conv2,512,5,strides=2,padding='same')
        conv3 = tf.layers.batch_normalization(conv3,training=True)
        conv3 = tf.layers.dropout(conv3,0.2)
        conv3 = tf.maximum(alpha * conv3, conv3)
        #shape is (4,4,512)
        
        layer = tf.reshape(conv3,(-1,4*4*512))
        logits = tf.layers.dense(layer,1)
        out = tf.sigmoid(logits)
        return logits, out


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

Generator¶

Implement generator to generate an image using z. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "generator" to allow the variables to be reused. The function should return the generated 28 x 28 x out_channel_dim images.

def generator(z, out_channel_dim, is_train=True, alpha=0.1):
    """ 
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    with tf.variable_scope('generator',reuse=(not is_train)):
        #z is 1d vector
        
        x1 = tf.layers.dense(z,2*2*1024)
        #x1 = tf.layers.dropout(x1, 0.2)
        x1 = tf.reshape(x1,(-1,2,2,1024))
        x1 = tf.layers.batch_normalization(x1,training=is_train)
        x1 = tf.layers.dropout(x1,0.2)
        x1 = tf.maximum(alpha * x1, x1)
        #shape is (2,2,1024)
        
        x2 = tf.layers.conv2d_transpose(x1, 512, 5, strides=2, padding='same')
        x2 = tf.layers.batch_normalization(x2,training=is_train)
        x2 = tf.layers.dropout(x2,0.2)
        x2 = tf.maximum(alpha * x2, x2)
        #shape is (4,4,512)
        
        x3 = tf.layers.conv2d_transpose(x2, 256, 4, strides=1, padding='valid')
        x3 = tf.layers.batch_normalization(x3,training=is_train)
        x3 = tf.layers.dropout(x3,0.2)
        x3 = tf.maximum(alpha * x3, x3)
        #shape is (7,7,256)
        
        x4 = tf.layers.conv2d_transpose(x3, 128, 5, strides=2, padding='same')
        x4 = tf.layers.batch_normalization(x4,training=is_train)
        x4 = tf.maximum(alpha * x4, x4)
        #shape is (14,14,512)
        
        logits = tf.layers.conv2d_transpose(x4, out_channel_dim, 5, strides=2, padding='same')
        
        out = tf.tanh(logits)
        return out


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

Loss¶

Implement model_loss to build the GANs for training and calculate the loss. The function should return a tuple of (discriminator loss, generator loss). Use the following functions you implemented:

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim, alpha=0.1):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
    g_out = generator(input_z,out_channel_dim,alpha=alpha,is_train=True)
    d_real_logits, d_real_out = discriminator(input_real, reuse=False, alpha=alpha)
    d_fake_logits, d_fake_out = discriminator(g_out, reuse=True, alpha=alpha)
    
    d_real_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_real_logits,labels=tf.ones_like(d_real_logits)))
    d_fake_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_fake_logits,labels=tf.zeros_like(d_fake_logits)))
    d_loss = d_real_loss + d_fake_loss
    g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_fake_logits,labels=tf.ones_like(d_fake_logits)))
    return d_loss, g_loss


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

Optimization¶

Implement model_opt to create the optimization operations for the GANs. Use tf.trainable_variables to get all the trainable variables. Filter the variables with names that are in the discriminator and generator scope names. The function should return a tuple of (discriminator training operation, generator training operation).

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """
    variables = tf.trainable_variables()
    d_variables = [var for var in variables if var.name.startswith('discriminator')]
    g_variables = [var for var in variables if var.name.startswith('generator')]
    
    with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
        d_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_variables)
        g_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(g_loss, var_list=g_variables)
    
    return d_opt, g_opt


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

Neural Network Training¶

Show Output¶

Use this function to show the current output of the generator during training. It will help you determine how well the GANs is training.

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

Train¶

Implement train to build and train the GANs. Use the following functions you implemented:

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

Use the show_generator_output to show generator output while you train. Running show_generator_output for every batch will drastically increase training time and increase the size of the notebook. It's recommended to print the generator output every 100 batches.

import time

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode, alpha = 0.1):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """
    
    image_channels=1 if data_image_mode=='L' else 3
    input_real, input_z, learningrate = model_inputs(data_shape[1],data_shape[2],image_channels,z_dim)
    d_loss,g_loss = model_loss(input_real,input_z,image_channels,alpha=alpha)
    d_opt, g_opt = model_opt(d_loss, g_loss, learningrate, beta1)
    
    saver = tf.train.Saver()
    losses = []
    
    start_time = time.time()
    
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count):
            steps=0
            for batch_images in get_batches(batch_size):
                steps+=1
                batch_images = batch_images *2 # change to -1 ~ 1
                
                batch_z = np.random.uniform(-1,1,size=(batch_size,z_dim))
                _ = sess.run(d_opt, feed_dict = {input_real:batch_images, input_z:batch_z, learningrate:learning_rate})
                _ = sess.run(g_opt, feed_dict = {input_real:batch_images,input_z:batch_z, learningrate:learning_rate})
                _ = sess.run(g_opt, feed_dict = {input_real:batch_images,input_z:batch_z, learningrate:learning_rate})
                
                if steps%10 == 0:
                    train_loss_d = d_loss.eval({input_z: batch_z, input_real: batch_images})
                    train_loss_g = g_loss.eval({input_z: batch_z})
                    print("Epoch {}/{}...".format(epoch_i+1, epoch_count),
                          "Batch {}/{}".format(steps,data_shape[0]/int(batch_size)),
                          "Discriminator Loss: {:.4f}...".format(train_loss_d),
                          "Generator Loss: {:.4f}".format(train_loss_g))
                    losses.append((train_loss_d, train_loss_g))
                    
                if steps%100 == 0:
                    show_generator_output(sess,25,input_z,image_channels,data_image_mode)
        saver.save(sess, './checkpoints/generator.ckpt')
        
        duration = time.time() - start_time
        
        return losses, duration

MNIST¶

Test your GANs architecture on MNIST. After 2 epochs, the GANs should be able to generate images that look like handwritten digits. Make sure the loss of the generator is lower than the loss of the discriminator or close to 0.

batch_size = 64
z_dim = 128
learning_rate = 0.003
beta1 = 0.6

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 2

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)

Epoch 1/2... Batch 10/937.5 Discriminator Loss: 1.4873... Generator Loss: 10.4977
Epoch 1/2... Batch 20/937.5 Discriminator Loss: 0.0789... Generator Loss: 11.5394
Epoch 1/2... Batch 30/937.5 Discriminator Loss: 0.9341... Generator Loss: 2.9322
Epoch 1/2... Batch 40/937.5 Discriminator Loss: 1.9763... Generator Loss: 0.9175
Epoch 1/2... Batch 50/937.5 Discriminator Loss: 2.1676... Generator Loss: 0.2520
Epoch 1/2... Batch 60/937.5 Discriminator Loss: 1.3444... Generator Loss: 1.0609
Epoch 1/2... Batch 70/937.5 Discriminator Loss: 2.9204... Generator Loss: 0.0693
Epoch 1/2... Batch 80/937.5 Discriminator Loss: 2.4457... Generator Loss: 0.7875
Epoch 1/2... Batch 90/937.5 Discriminator Loss: 4.1518... Generator Loss: 0.0234
Epoch 1/2... Batch 100/937.5 Discriminator Loss: 0.9666... Generator Loss: 1.8574

Epoch 1/2... Batch 110/937.5 Discriminator Loss: 1.5912... Generator Loss: 1.5546
Epoch 1/2... Batch 120/937.5 Discriminator Loss: 1.6894... Generator Loss: 9.0620
Epoch 1/2... Batch 130/937.5 Discriminator Loss: 1.0768... Generator Loss: 0.5673
Epoch 1/2... Batch 140/937.5 Discriminator Loss: 2.2680... Generator Loss: 1.3141
Epoch 1/2... Batch 150/937.5 Discriminator Loss: 1.2830... Generator Loss: 9.6682
Epoch 1/2... Batch 160/937.5 Discriminator Loss: 1.0028... Generator Loss: 1.5470
Epoch 1/2... Batch 170/937.5 Discriminator Loss: 0.3539... Generator Loss: 2.5996
Epoch 1/2... Batch 180/937.5 Discriminator Loss: 1.4464... Generator Loss: 0.7808
Epoch 1/2... Batch 190/937.5 Discriminator Loss: 2.9499... Generator Loss: 0.2498
Epoch 1/2... Batch 200/937.5 Discriminator Loss: 2.0959... Generator Loss: 0.2851

Epoch 1/2... Batch 210/937.5 Discriminator Loss: 2.0707... Generator Loss: 1.0712
Epoch 1/2... Batch 220/937.5 Discriminator Loss: 0.2891... Generator Loss: 1.8029
Epoch 1/2... Batch 230/937.5 Discriminator Loss: 0.2742... Generator Loss: 4.3543
Epoch 1/2... Batch 240/937.5 Discriminator Loss: 0.7259... Generator Loss: 0.8364
Epoch 1/2... Batch 250/937.5 Discriminator Loss: 0.9043... Generator Loss: 5.7396
Epoch 1/2... Batch 260/937.5 Discriminator Loss: 2.9697... Generator Loss: 3.5579
Epoch 1/2... Batch 270/937.5 Discriminator Loss: 0.5121... Generator Loss: 1.5424
Epoch 1/2... Batch 280/937.5 Discriminator Loss: 3.2084... Generator Loss: 0.0955
Epoch 1/2... Batch 290/937.5 Discriminator Loss: 0.7774... Generator Loss: 4.0743
Epoch 1/2... Batch 300/937.5 Discriminator Loss: 0.4631... Generator Loss: 5.0264

Epoch 1/2... Batch 310/937.5 Discriminator Loss: 0.9330... Generator Loss: 2.3301
Epoch 1/2... Batch 320/937.5 Discriminator Loss: 1.5297... Generator Loss: 1.7190
Epoch 1/2... Batch 330/937.5 Discriminator Loss: 0.8513... Generator Loss: 1.5190
Epoch 1/2... Batch 340/937.5 Discriminator Loss: 0.8325... Generator Loss: 2.4536
Epoch 1/2... Batch 350/937.5 Discriminator Loss: 0.7937... Generator Loss: 1.4989
Epoch 1/2... Batch 360/937.5 Discriminator Loss: 1.1520... Generator Loss: 1.3396
Epoch 1/2... Batch 370/937.5 Discriminator Loss: 0.8359... Generator Loss: 0.9807
Epoch 1/2... Batch 380/937.5 Discriminator Loss: 1.0989... Generator Loss: 1.6899
Epoch 1/2... Batch 390/937.5 Discriminator Loss: 1.3080... Generator Loss: 0.5272
Epoch 1/2... Batch 400/937.5 Discriminator Loss: 0.7330... Generator Loss: 1.2535

Epoch 1/2... Batch 410/937.5 Discriminator Loss: 0.7898... Generator Loss: 2.0798
Epoch 1/2... Batch 420/937.5 Discriminator Loss: 2.2841... Generator Loss: 0.1873
Epoch 1/2... Batch 430/937.5 Discriminator Loss: 0.8764... Generator Loss: 0.9488
Epoch 1/2... Batch 440/937.5 Discriminator Loss: 2.1357... Generator Loss: 0.2803
Epoch 1/2... Batch 450/937.5 Discriminator Loss: 2.6818... Generator Loss: 1.5785
Epoch 1/2... Batch 460/937.5 Discriminator Loss: 1.1709... Generator Loss: 1.9969
Epoch 1/2... Batch 470/937.5 Discriminator Loss: 0.6695... Generator Loss: 1.3704
Epoch 1/2... Batch 480/937.5 Discriminator Loss: 0.7698... Generator Loss: 0.9434
Epoch 1/2... Batch 490/937.5 Discriminator Loss: 1.4701... Generator Loss: 0.4148
Epoch 1/2... Batch 500/937.5 Discriminator Loss: 0.7576... Generator Loss: 1.1842

Epoch 1/2... Batch 510/937.5 Discriminator Loss: 0.9066... Generator Loss: 0.9441
Epoch 1/2... Batch 520/937.5 Discriminator Loss: 1.4286... Generator Loss: 0.4845
Epoch 1/2... Batch 530/937.5 Discriminator Loss: 1.1102... Generator Loss: 2.2303
Epoch 1/2... Batch 540/937.5 Discriminator Loss: 1.2433... Generator Loss: 0.5920
Epoch 1/2... Batch 550/937.5 Discriminator Loss: 1.4153... Generator Loss: 1.7773
Epoch 1/2... Batch 560/937.5 Discriminator Loss: 1.0957... Generator Loss: 2.2232
Epoch 1/2... Batch 570/937.5 Discriminator Loss: 1.2898... Generator Loss: 1.2640
Epoch 1/2... Batch 580/937.5 Discriminator Loss: 1.1635... Generator Loss: 1.2399
Epoch 1/2... Batch 590/937.5 Discriminator Loss: 1.0245... Generator Loss: 0.8148
Epoch 1/2... Batch 600/937.5 Discriminator Loss: 1.2484... Generator Loss: 0.6608

Epoch 1/2... Batch 610/937.5 Discriminator Loss: 0.8211... Generator Loss: 0.9220
Epoch 1/2... Batch 620/937.5 Discriminator Loss: 1.3116... Generator Loss: 1.1024
Epoch 1/2... Batch 630/937.5 Discriminator Loss: 1.0454... Generator Loss: 1.8178
Epoch 1/2... Batch 640/937.5 Discriminator Loss: 1.4739... Generator Loss: 0.4555
Epoch 1/2... Batch 650/937.5 Discriminator Loss: 1.0765... Generator Loss: 0.8824
Epoch 1/2... Batch 660/937.5 Discriminator Loss: 1.7769... Generator Loss: 2.1706
Epoch 1/2... Batch 670/937.5 Discriminator Loss: 1.1123... Generator Loss: 0.7004
Epoch 1/2... Batch 680/937.5 Discriminator Loss: 1.9762... Generator Loss: 0.2004
Epoch 1/2... Batch 690/937.5 Discriminator Loss: 1.4269... Generator Loss: 1.0431
Epoch 1/2... Batch 700/937.5 Discriminator Loss: 0.9614... Generator Loss: 0.6754

Epoch 1/2... Batch 710/937.5 Discriminator Loss: 0.7843... Generator Loss: 1.3666
Epoch 1/2... Batch 720/937.5 Discriminator Loss: 1.2292... Generator Loss: 0.7885
Epoch 1/2... Batch 730/937.5 Discriminator Loss: 1.4557... Generator Loss: 0.4793
Epoch 1/2... Batch 740/937.5 Discriminator Loss: 1.2728... Generator Loss: 0.8177
Epoch 1/2... Batch 750/937.5 Discriminator Loss: 1.0093... Generator Loss: 1.9030
Epoch 1/2... Batch 760/937.5 Discriminator Loss: 1.0415... Generator Loss: 0.7915
Epoch 1/2... Batch 770/937.5 Discriminator Loss: 1.0996... Generator Loss: 1.4395
Epoch 1/2... Batch 780/937.5 Discriminator Loss: 0.8402... Generator Loss: 1.0362
Epoch 1/2... Batch 790/937.5 Discriminator Loss: 1.4281... Generator Loss: 2.2084
Epoch 1/2... Batch 800/937.5 Discriminator Loss: 1.2055... Generator Loss: 0.6637

Epoch 1/2... Batch 810/937.5 Discriminator Loss: 1.4896... Generator Loss: 0.3970
Epoch 1/2... Batch 820/937.5 Discriminator Loss: 1.2586... Generator Loss: 2.0728
Epoch 1/2... Batch 830/937.5 Discriminator Loss: 1.1488... Generator Loss: 1.8982
Epoch 1/2... Batch 840/937.5 Discriminator Loss: 1.4225... Generator Loss: 0.4159
Epoch 1/2... Batch 850/937.5 Discriminator Loss: 0.8088... Generator Loss: 1.5711
Epoch 1/2... Batch 860/937.5 Discriminator Loss: 2.5942... Generator Loss: 0.1135
Epoch 1/2... Batch 870/937.5 Discriminator Loss: 1.2077... Generator Loss: 0.7404
Epoch 1/2... Batch 880/937.5 Discriminator Loss: 1.0890... Generator Loss: 1.7079
Epoch 1/2... Batch 890/937.5 Discriminator Loss: 1.6620... Generator Loss: 0.3674
Epoch 1/2... Batch 900/937.5 Discriminator Loss: 1.7578... Generator Loss: 0.2661

Epoch 1/2... Batch 910/937.5 Discriminator Loss: 1.2208... Generator Loss: 0.8694
Epoch 1/2... Batch 920/937.5 Discriminator Loss: 1.6735... Generator Loss: 0.3131
Epoch 1/2... Batch 930/937.5 Discriminator Loss: 1.1971... Generator Loss: 0.5442
Epoch 2/2... Batch 10/937.5 Discriminator Loss: 1.1125... Generator Loss: 2.2360
Epoch 2/2... Batch 20/937.5 Discriminator Loss: 1.3706... Generator Loss: 0.4363
Epoch 2/2... Batch 30/937.5 Discriminator Loss: 0.7306... Generator Loss: 1.4356
Epoch 2/2... Batch 40/937.5 Discriminator Loss: 1.2296... Generator Loss: 0.5706
Epoch 2/2... Batch 50/937.5 Discriminator Loss: 1.3910... Generator Loss: 2.2709
Epoch 2/2... Batch 60/937.5 Discriminator Loss: 1.4075... Generator Loss: 2.0493
Epoch 2/2... Batch 70/937.5 Discriminator Loss: 1.3298... Generator Loss: 1.3974
Epoch 2/2... Batch 80/937.5 Discriminator Loss: 2.4190... Generator Loss: 3.3017
Epoch 2/2... Batch 90/937.5 Discriminator Loss: 1.1612... Generator Loss: 1.4789
Epoch 2/2... Batch 100/937.5 Discriminator Loss: 1.0818... Generator Loss: 0.7688

CelebA¶

Run your GANs on CelebA. It will take around 20 minutes on the average GPU to run one epoch. You can run the whole epoch or stop when it starts to generate realistic faces.

batch_size = 16
z_dim = 256
learning_rate = 0.00005
beta1 = 0.6


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 1

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)

Epoch 1/1... Batch 10/12662.4375 Discriminator Loss: 4.1553... Generator Loss: 0.0267
Epoch 1/1... Batch 20/12662.4375 Discriminator Loss: 3.4026... Generator Loss: 0.0919
Epoch 1/1... Batch 30/12662.4375 Discriminator Loss: 3.5148... Generator Loss: 0.2356
Epoch 1/1... Batch 40/12662.4375 Discriminator Loss: 1.4051... Generator Loss: 0.8152
Epoch 1/1... Batch 50/12662.4375 Discriminator Loss: 1.3392... Generator Loss: 0.7356
Epoch 1/1... Batch 60/12662.4375 Discriminator Loss: 1.4812... Generator Loss: 0.8031
Epoch 1/1... Batch 70/12662.4375 Discriminator Loss: 1.2973... Generator Loss: 1.0091
Epoch 1/1... Batch 80/12662.4375 Discriminator Loss: 1.5198... Generator Loss: 0.9252
Epoch 1/1... Batch 90/12662.4375 Discriminator Loss: 0.8775... Generator Loss: 1.3702
Epoch 1/1... Batch 100/12662.4375 Discriminator Loss: 0.4515... Generator Loss: 1.7539

Epoch 1/1... Batch 110/12662.4375 Discriminator Loss: 1.1833... Generator Loss: 1.0037
Epoch 1/1... Batch 120/12662.4375 Discriminator Loss: 1.1128... Generator Loss: 1.4329
Epoch 1/1... Batch 130/12662.4375 Discriminator Loss: 1.0697... Generator Loss: 1.2092
Epoch 1/1... Batch 140/12662.4375 Discriminator Loss: 1.1013... Generator Loss: 1.3225
Epoch 1/1... Batch 150/12662.4375 Discriminator Loss: 0.6902... Generator Loss: 1.6279
Epoch 1/1... Batch 160/12662.4375 Discriminator Loss: 0.3558... Generator Loss: 1.6022
Epoch 1/1... Batch 170/12662.4375 Discriminator Loss: 0.7766... Generator Loss: 1.3247
Epoch 1/1... Batch 180/12662.4375 Discriminator Loss: 0.4001... Generator Loss: 1.7013
Epoch 1/1... Batch 190/12662.4375 Discriminator Loss: 1.2543... Generator Loss: 1.3668
Epoch 1/1... Batch 200/12662.4375 Discriminator Loss: 0.7212... Generator Loss: 1.3482

Epoch 1/1... Batch 210/12662.4375 Discriminator Loss: 0.3984... Generator Loss: 1.7717
Epoch 1/1... Batch 220/12662.4375 Discriminator Loss: 1.2423... Generator Loss: 1.3485
Epoch 1/1... Batch 230/12662.4375 Discriminator Loss: 1.2410... Generator Loss: 1.4070
Epoch 1/1... Batch 240/12662.4375 Discriminator Loss: 0.2451... Generator Loss: 2.1051
Epoch 1/1... Batch 250/12662.4375 Discriminator Loss: 0.7483... Generator Loss: 1.3146
Epoch 1/1... Batch 260/12662.4375 Discriminator Loss: 0.4546... Generator Loss: 1.7107
Epoch 1/1... Batch 270/12662.4375 Discriminator Loss: 0.3890... Generator Loss: 1.8344
Epoch 1/1... Batch 280/12662.4375 Discriminator Loss: 0.3948... Generator Loss: 1.3797
Epoch 1/1... Batch 290/12662.4375 Discriminator Loss: 0.2170... Generator Loss: 2.1330
Epoch 1/1... Batch 300/12662.4375 Discriminator Loss: 0.4483... Generator Loss: 1.9972

Epoch 1/1... Batch 310/12662.4375 Discriminator Loss: 0.2546... Generator Loss: 1.9341
Epoch 1/1... Batch 320/12662.4375 Discriminator Loss: 0.8105... Generator Loss: 1.2810
Epoch 1/1... Batch 330/12662.4375 Discriminator Loss: 0.3410... Generator Loss: 2.1792
Epoch 1/1... Batch 340/12662.4375 Discriminator Loss: 0.9294... Generator Loss: 1.6609
Epoch 1/1... Batch 350/12662.4375 Discriminator Loss: 0.6728... Generator Loss: 1.8682
Epoch 1/1... Batch 360/12662.4375 Discriminator Loss: 0.8441... Generator Loss: 1.2790
Epoch 1/1... Batch 370/12662.4375 Discriminator Loss: 0.2281... Generator Loss: 2.0811
Epoch 1/1... Batch 380/12662.4375 Discriminator Loss: 0.1314... Generator Loss: 2.7254
Epoch 1/1... Batch 390/12662.4375 Discriminator Loss: 0.3516... Generator Loss: 2.2161
Epoch 1/1... Batch 400/12662.4375 Discriminator Loss: 0.1108... Generator Loss: 2.4534

Epoch 1/1... Batch 410/12662.4375 Discriminator Loss: 0.3449... Generator Loss: 1.9220
Epoch 1/1... Batch 420/12662.4375 Discriminator Loss: 0.1938... Generator Loss: 1.9523
Epoch 1/1... Batch 430/12662.4375 Discriminator Loss: 0.1740... Generator Loss: 2.3455
Epoch 1/1... Batch 440/12662.4375 Discriminator Loss: 0.2618... Generator Loss: 2.3098
Epoch 1/1... Batch 450/12662.4375 Discriminator Loss: 0.1759... Generator Loss: 2.7148
Epoch 1/1... Batch 460/12662.4375 Discriminator Loss: 0.2606... Generator Loss: 2.2316
Epoch 1/1... Batch 470/12662.4375 Discriminator Loss: 0.1355... Generator Loss: 2.2792
Epoch 1/1... Batch 480/12662.4375 Discriminator Loss: 0.1317... Generator Loss: 2.3862
Epoch 1/1... Batch 490/12662.4375 Discriminator Loss: 0.1835... Generator Loss: 2.6702
Epoch 1/1... Batch 500/12662.4375 Discriminator Loss: 0.3227... Generator Loss: 2.1365

Epoch 1/1... Batch 510/12662.4375 Discriminator Loss: 0.4367... Generator Loss: 2.3112
Epoch 1/1... Batch 520/12662.4375 Discriminator Loss: 0.2095... Generator Loss: 2.4195
Epoch 1/1... Batch 530/12662.4375 Discriminator Loss: 0.2084... Generator Loss: 2.3388
Epoch 1/1... Batch 540/12662.4375 Discriminator Loss: 0.1060... Generator Loss: 2.7414
Epoch 1/1... Batch 550/12662.4375 Discriminator Loss: 0.1801... Generator Loss: 2.2564
Epoch 1/1... Batch 560/12662.4375 Discriminator Loss: 0.2015... Generator Loss: 2.6514
Epoch 1/1... Batch 570/12662.4375 Discriminator Loss: 0.2350... Generator Loss: 2.1058
Epoch 1/1... Batch 580/12662.4375 Discriminator Loss: 0.0701... Generator Loss: 3.0417
Epoch 1/1... Batch 590/12662.4375 Discriminator Loss: 0.0865... Generator Loss: 3.1887
Epoch 1/1... Batch 600/12662.4375 Discriminator Loss: 0.2489... Generator Loss: 2.4069

Epoch 1/1... Batch 610/12662.4375 Discriminator Loss: 0.1488... Generator Loss: 2.8781
Epoch 1/1... Batch 620/12662.4375 Discriminator Loss: 1.4375... Generator Loss: 0.4347
Epoch 1/1... Batch 630/12662.4375 Discriminator Loss: 1.8970... Generator Loss: 0.4759
Epoch 1/1... Batch 640/12662.4375 Discriminator Loss: 1.5362... Generator Loss: 0.6808
Epoch 1/1... Batch 650/12662.4375 Discriminator Loss: 2.0841... Generator Loss: 0.4849
Epoch 1/1... Batch 660/12662.4375 Discriminator Loss: 1.1454... Generator Loss: 0.7964
Epoch 1/1... Batch 670/12662.4375 Discriminator Loss: 1.3317... Generator Loss: 0.8355
Epoch 1/1... Batch 680/12662.4375 Discriminator Loss: 2.1379... Generator Loss: 0.4303
Epoch 1/1... Batch 690/12662.4375 Discriminator Loss: 1.0464... Generator Loss: 1.0675
Epoch 1/1... Batch 700/12662.4375 Discriminator Loss: 1.5479... Generator Loss: 0.6609

Epoch 1/1... Batch 710/12662.4375 Discriminator Loss: 0.9563... Generator Loss: 0.9273
Epoch 1/1... Batch 720/12662.4375 Discriminator Loss: 1.1546... Generator Loss: 0.7694
Epoch 1/1... Batch 730/12662.4375 Discriminator Loss: 1.2550... Generator Loss: 0.7900
Epoch 1/1... Batch 740/12662.4375 Discriminator Loss: 0.9704... Generator Loss: 1.1757
Epoch 1/1... Batch 750/12662.4375 Discriminator Loss: 0.9481... Generator Loss: 1.1402
Epoch 1/1... Batch 760/12662.4375 Discriminator Loss: 1.2228... Generator Loss: 1.1402
Epoch 1/1... Batch 770/12662.4375 Discriminator Loss: 1.2264... Generator Loss: 0.8511
Epoch 1/1... Batch 780/12662.4375 Discriminator Loss: 1.1557... Generator Loss: 0.8020
Epoch 1/1... Batch 790/12662.4375 Discriminator Loss: 1.5980... Generator Loss: 0.5880
Epoch 1/1... Batch 800/12662.4375 Discriminator Loss: 1.5015... Generator Loss: 0.6012

Epoch 1/1... Batch 810/12662.4375 Discriminator Loss: 2.3580... Generator Loss: 0.2278
Epoch 1/1... Batch 820/12662.4375 Discriminator Loss: 1.7282... Generator Loss: 0.6127
Epoch 1/1... Batch 830/12662.4375 Discriminator Loss: 1.1424... Generator Loss: 0.7692
Epoch 1/1... Batch 840/12662.4375 Discriminator Loss: 1.3940... Generator Loss: 0.7166
Epoch 1/1... Batch 850/12662.4375 Discriminator Loss: 1.5581... Generator Loss: 0.5661
Epoch 1/1... Batch 860/12662.4375 Discriminator Loss: 1.7214... Generator Loss: 0.7255
Epoch 1/1... Batch 870/12662.4375 Discriminator Loss: 1.1094... Generator Loss: 1.0418
Epoch 1/1... Batch 880/12662.4375 Discriminator Loss: 0.9299... Generator Loss: 1.1405
Epoch 1/1... Batch 890/12662.4375 Discriminator Loss: 1.2304... Generator Loss: 0.7446
Epoch 1/1... Batch 900/12662.4375 Discriminator Loss: 1.0265... Generator Loss: 0.8371

Epoch 1/1... Batch 910/12662.4375 Discriminator Loss: 1.2947... Generator Loss: 1.1008
Epoch 1/1... Batch 920/12662.4375 Discriminator Loss: 1.5716... Generator Loss: 0.3940
Epoch 1/1... Batch 930/12662.4375 Discriminator Loss: 1.1647... Generator Loss: 0.9638
Epoch 1/1... Batch 940/12662.4375 Discriminator Loss: 1.4818... Generator Loss: 0.8343
Epoch 1/1... Batch 950/12662.4375 Discriminator Loss: 0.8868... Generator Loss: 1.1776
Epoch 1/1... Batch 960/12662.4375 Discriminator Loss: 1.3705... Generator Loss: 0.7533
Epoch 1/1... Batch 970/12662.4375 Discriminator Loss: 1.3771... Generator Loss: 0.6388
Epoch 1/1... Batch 980/12662.4375 Discriminator Loss: 2.0860... Generator Loss: 0.3850
Epoch 1/1... Batch 990/12662.4375 Discriminator Loss: 1.0318... Generator Loss: 0.6668
Epoch 1/1... Batch 1000/12662.4375 Discriminator Loss: 0.8758... Generator Loss: 1.1389

Submitting This Project¶

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.